Celia Quillan
speaker
53 appearances
1 recordings
1 series
first heard Mar 2025
last heard Mar 2025
Celia Quillan’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsNo recordings in the last 12 months.Older appearances are listed below; set an alert to hear about the next one.
Appearances
And in its training data, it's associated, you know, snappy characters in a text it might've been trained off of, right? Maybe a character is described as snappy in their dialogue at some point in time. It starts to associate those words with different kinds of output.
So it's hard to explain it exactly, but it does have a very advanced way of understanding human language patterns and human language context and meaning. So if you ask it to be snappier, it will probably develop a response back to you that is more concise, maybe a little bit sassy. Maybe it has a rhyme to it, depending on what else is in your prompt.
But it because it has become like an expert of mimicking human language patterns, it effectively is very good at understanding the meanings of words that you use. Now, if you use the gibberish word that had never existed before and said, can you make it more smorgasbord?
It would have no idea what you meant, but it might come up with something that's, you know, similarly silly sounding as smorgasbord.
Yes, that's a good way to put it. It is definitely attempt to. You'll see this with the earlier models from ChatGPT. These tools are slowly getting better at it, but if you ask it to, for example, make something funny, make it funnier, it will have a little bit more trouble with that.
It will kind of give you a response back that sounds like it could be a premise for a joke, but then it does the punchline falls flat. And that's an important thing to understand about these tools is they don't have the same, you know, human experience that we have. They don't really know anything. They're just guessing at the best answer.
So I believe at some point I asked it to tell me a knock, knock joke about the dentist. And it was like, knock, knock. Who's there? Dentist. dentist to cavity another. And it, you know, that doesn't make any sense to us, but it was in the structure of a joke.
So, so, but for the most part, if there is, you know, I think humor is the one that really falls flat on, but cleverness, you know, if you ask it to create 10 clever headlines, maybe one of them, you pluck it out after you read it through as a human is clever. The other nine might not be so much, but one of those 10 is bound to get you a little closer to where you want to be.
That is a super good question. And the answer is you should be always a little skeptical with everything, especially if it's a super niche topic because these tools do create, you know, their best, they're just predicting the next best answer. That prediction could be wrong. Just kind of like how the weather prediction might be wrong for the day. If you're looking at your weather app. But yeah,
these tools are progressively getting better and better and trained on more data sets, which are making them less likely to what we call hallucinate. Mathematics, they used to be terrible at predicting the best answer to a mathematical question. Now, the more advanced models, especially those that are,
are called deep reasoning models where they kind of, when they generate a response to you, it takes a little bit longer because it's actually running through its response through itself over and over again to see if it's really getting to the best answer. Those will be more likely to be correct.
Now, when it comes to general questions about things that are likely, there's likely a lot of information about it on the internet, for example, therefore it's likely trained on a lot of that data. Like why is the sky blue? it's probably going to give a pretty solid answer to that.
If you ask it about a very niche disease that just appeared in the world that hasn't been heavily researched, it's probably going to give you an answer, but it may not be correct. The best thing I tell people to do is that when you use these tools as a jumping off point, because they can get you really far, and then do a quick Google search, see if there's a secondary source that backs that up.
You'll find that a lot of the time, it is totally right. It's wrong enough of the time that you should always do that extra step of fact-checking.
You will get one lasagna recipe back. It will not necessarily be the best lasagna. Best is, you know, arbitrary. But you will get a recipe back that is most likely to result in a lasagna. That's the best way I can put it. So if you truly want to have the vetted, you know, best ranked lasagna, lots of great ratings, Google might still be the best way to go.
But if you're looking to just get started, I want to try and cook something quick. I do this all the time. How do I? Give me a quick recipe. I have chicken, asparagus. Parmesan and tomato sauce in my fridge. What can I make with this? It will generate back a recipe that will result in something tasty. It may not be the best thing you've ever made, but it will get you there.
And that is again, because it's taking in all of the data it has of all the lasagna recipes that are in its training data, which is likely a massive amount because they are effectively trained on the whole variety of things on the internet.
That's correct. You ask for one, you get one. Now you can ask for five and it'll give you five different ones. I do say in the case of a recipe, if you're looking for a recipe, most of these tools have a certain text-based cutoff. After they generate a certain number of words, they'll finish the answer. That way they don't go on for forever and you'll have 10 pages of content to read.
So if you're looking for recipes, for example, sometimes you're better off asking for one and then saying, okay, that sounds okay, but could I have another? And then it generates another one. But I like it, honestly.
Instead of having to scroll through endless pages of search results and then click through them, scroll past all of the ads and all of the recipe blogs tend to have a lot of extra context up front that has nothing to do with the recipe. ChatGPT will give you exactly what you want, nicely packaged, short and sweet.
Showing 21–40 of 53 · page 2 of 3
← Previous
Next →